Why a video made by an AI agent should stay editable

When an AI agent makes a video in SPREVA, it makes a Studio project: scenes on a timeline, a script, a voice track, pictures and captions, each of them separate. A change is a change to one of those parts, not a new video, so you can ask for “the second scene shorter” and get exactly that, with everything else where it was. Each change is one step you can undo.

Most video generators end with a file. You describe what you want, wait, and get an MP4. If it is close but not right, you describe it again and wait again, and the new file is not the old one with a fix: the pacing has moved, the voice reads a little differently, the picture you liked in scene three is gone. You are not editing. You are rolling dice with a longer prompt.

For a one-off that can be fine. For social media it is not, because the work is never one video. It is the same video shortened for TikTok, reframed for a Story, with a different opening for LinkedIn, and then a new one next week that should look like this one. A file cannot carry any of that forward.

Editable does not mean a text box where you paste a new prompt. It means the thing the agent made has parts, and each part can be changed on its own:

• The script is text. Change a line and the narration can be recorded again, while the pictures and the captions keep their places.
• The pictures sit on the words they belong to. Swap a photo and the captions, the voice and the other scenes do not move.
• Every change is one step. Asked for by you or made by the agent, it goes into the same history, and undo takes it back.
• The timeline is yours too. Anything the agent can do you can do by hand: trim a scene, restyle the captions, add a logo, and the agent sees the result the next time you ask.

In practice it is a conversation. With your agent connected to SPREVA, you ask for the video the way you would ask a colleague:

The agent starts a project, writes the script, asks before the voice-over spends AI credits, places each photo on the line that describes it, checks the cut and exports it. Then you look at it, and you ask again:

Only the opening and one picture change. The voice, the captions and the other scenes stay exactly where they were, because the agent is editing the project, not making a new one. If you would rather drag the scene yourself, open the timeline: it is the same project, and the agent's next change lands on top of yours.

When a person edits, they see what they are doing. When an agent edits, you see the result, and you need to trust that the result is what you asked for and nothing more. An agent that answers “shorter opening” with a new video has not done what you asked; it has done something that includes what you asked. The only way to tell the difference is to watch the whole thing again.

A project fixes this. The agent says what it changed, in scenes and seconds, and the history shows the same. You review the change, not the video. That is what makes it possible to hand over the repetitive work: the next ten versions for the next ten networks, the weekly video that should look like last week's, the caption in Spanish. Each is a change to something you already approved.

Upload a handful of photos, ask your agent for a short vertical video, and then ask for one thing to be different. Watch what stays the same. Step by step, in the app or from ChatGPT or Claude, with Studio in the Pro and Business plans and a 7-day trial on both.